Mao Yingjie sent the pattern scan results on a Friday.
Not a summary. Twelve pages of documented probability vectors — each one a directional force in the city's ambient probability field, mapped against the existing case data and the mechanism activity logs she had been building since the Monday meeting.
He read it twice. Then he called her.
"Walk me through it," he said.
"The dominant pattern is in the financial sector." Her voice had the quality it got when she'd been running the pattern recognition for extended hours — precise, slightly flat, like she'd stripped everything nonessential. "There's a high-probability instability event building in one of the Capital Alliance's mid-tier member organizations. The pattern suggests a leadership crisis — two competing internal factions, both accumulating resources and ally commitments, on a collision trajectory." She paused. "The collision is six to eight weeks out."
"Which organization?"
"Heming Capital. They're mid-tier financial, affiliated with three of the city's second-rank awakened factions. The power world interest is indirect — but if Heming collapses under an internal split, the three affiliated factions lose their primary financial infrastructure simultaneously."
He had a working model of Heming Capital from the background intelligence he maintained. Mid-tier, yes, but mid-tier in the Capital Alliance's financial structure meant: significant. Destabilizing three affiliated power factions simultaneously was not a mid-tier event.
"The anchor," he said. "What does Li Xiulan need to do?"
"The pattern is still in the building phase," Mao Yingjie said. "The two internal factions haven't fully committed to the collision — there are decision nodes over the next two weeks where either side could pull back. What the anchor needs to do is reduce the variance around those decision nodes. Make the pull-back options more probable relative to the escalation options."
"She needs to hold the field stable while I design the intervention," he said.
"Yes. Your intervention at the right decision node would shift the trajectory without requiring visible action. But the intervention needs the ambient probability distribution to be stable enough that the shift registers as decisive rather than incremental."
He sat with this. The coordinated task architecture — exactly what the system had described. Mao Yingjie identifies the pattern. Li Xiulan anchors the field. He designs the intervention. Three steps in sequence, each one dependent on the previous.
"Forward it to Li Xiulan," he said. "She needs the vector analysis."
"Already sent," she said. "Two hours ago. I waited to see if you'd find the problem I found."
He looked at the twelve pages again. "What problem."
"Page nine," she said. "The probability vector running from the Heming internal faction structure into the Lin Family Faction's financial network." A pause. "Lin Zhengyue has an active financial position in Heming Capital. If the Heming crisis develops, it affects her."
He turned to page nine.
The vector was there. A light connection — not a primary financial dependency, but a secondary exposure through the Lin Family's infrastructure investments. If Heming destabilized severely, the Lin Family would see financial pressure in the same 6-8 week window.
The same window as the Capital Alliance expansion meeting.
He looked at the timeline.
He sat back.
---
Li Xiulan called at noon.
"The vector analysis," she said. "The anchor work you need — I can do it. But there's something wrong with the model you gave me."
"Go ahead," he said.
"The coordination task assumes my anchor field operates uniformly across the target probability distribution," she said. "It doesn't. The field is object-specific — I don't reduce variance across an entire environmental distribution. I reduce variance around specific identified objectives. One objective at a time, prioritized by what my mechanism reads as critical."
He stopped.
He had been modeling her mechanism as ambient field regulation. Diffuse variance reduction across a spatial area. That was what the system's description had initially suggested. Even after the Monday meeting, when he had updated the model to "active navigator" — he had maintained the assumption that the navigation operated across a broad field.
It didn't.
"You work sequentially," he said. "Not simultaneously."
"Yes," she said. "When I identify an objective as critical, the anchor field targets it specifically. Everything else in the distribution is unaffected." She paused. "If the Heming intervention requires me to stabilize multiple decision nodes simultaneously—"
"You can't do it simultaneously," he said.
"No. I can do them sequentially. But the sequence has to be correct, and the timing between nodes matters." A pause. "If node two is the decision window and node one is still in progress when node two opens—"
"The intervention fails at node two," he said. "And the field at node one is already committed."
"Yes."
He looked at the twelve-page document. Mao Yingjie's analysis had identified seven decision nodes in the two-week window. Seven points where the Heming internal factions could pull back or escalate. He had been planning a single intervention timed to the highest-probability node — assuming Li Xiulan's field would cover the surrounding distribution.
It wouldn't. He'd been wrong about the mechanism for a week and a half. The coordination plan he had been building was built on a wrong model.
"When did you know this?" he said.
"When I read your preliminary plan outline," she said. "Two days ago."
"Why didn't you flag it earlier?"
A short pause. "I was checking whether the model correction would require rebuilding from scratch or whether it would produce a modified version that worked with the corrected assumptions." She paused. "It requires rebuilding."
He put the document down.
The coordinated task. Fourteen days. Six days already elapsed. The remaining eight days were sufficient for a rebuilt plan — but barely. And the rebuild required going back to Mao Yingjie for a reordered analysis: which decision nodes were most critical, which sequence would maximize Li Xiulan's anchor effectiveness given the sequential constraint, and whether the intervention point he had been targeting was still correct.
"I needed three more days to develop the full picture before I corrected you," Li Xiulan said. "That was the wrong decision. I was prioritizing the quality of my correction over the time cost of the delay."
She had just told him that she had made a mistake. Without framing it differently.
"Yes," he said. "Flag it immediately next time."
"Understood," she said.
"Where are you now on the anchor work?"
"Node three," she said. "The third decision window in Mao Yingjie's timeline. I started there because my mechanism identified it as the highest-criticality point. The field is stable." She paused. "The field will hold at node three for approximately four days before the pressure from the adjacent distribution overrides it. You have four days to design the intervention."
Four days. With a wrong plan that needed rebuilding.
He called Mao Yingjie back.
---
The rebuild took fourteen hours across two days.
Mao Yingjie ran the reanalysis with the corrected constraint: seven decision nodes, Li Xiulan's anchor operating sequentially, priority order to be determined by network analysis of which nodes were genuinely decisive versus which were noise. She identified three primary nodes — the rest were secondary, outcomes that would follow the primary decisions.
Li Xiulan reoriented from node three (secondary, it turned out) to node two (primary). The anchor field moved. The timeline reset.
Chen rebuilt the intervention plan around the primary node sequence.
The intervention itself was not combat, not confrontation. It was information. The Heming Capital internal conflict was running on incomplete information — the leadership faction underestimated the resources the challenging faction had assembled, the challenging faction underestimated the institutional loyalty the leadership held. Both sides were escalating toward a collision that neither of them would win cleanly because neither of them could see the complete picture.
A correctly timed information correction — delivered to the right person in each faction, through the right channel — would recalibrate both sides' assessment simultaneously. The collision trajectory would soften. The pull-back option would register as viable.
He had the channels. He had the information. He had the Vanguard's analytical infrastructure and two years of building city-level intelligence sources.
What he hadn't had was the correct model of Li Xiulan's mechanism.
He thought about this on the second day, an hour of it, while Mao Yingjie's reanalysis was running and Li Xiulan was holding node two's field stability. He had made a working assumption, hadn't tested it directly, and built two weeks of planning on an unchecked foundation. The plan had been logical on its own terms. The terms were wrong.
He noted this in the category it belonged to: errors that come from not asking the question you should have asked on day one.
*How, exactly, does your mechanism work?* He had heard Li Xiulan describe it. He had seen her use it. He had built a model that felt consistent with the data. He had not asked: *Is this model complete?*
He added the question to his standard verification protocol. Not as a punishment — as a correction to a gap.
---
The intervention executed on day four of the rebuilt plan.
The information correction reached both Heming factions within a three-hour window, through two separate channels that had no visible connection to each other. The leadership faction received updated data on the challenger's resource position — accurate, no embellishment, delivered through a financial intelligence contact that the leadership trusted. The challenging faction received updated data on their leadership's institutional loyalty numbers — accurate, same standard.
Both factions recalibrated.
The collision trajectory, per Mao Yingjie's real-time monitoring, shifted. The two internal groups pulled back to a negotiated settlement conversation instead of the resource-commitment escalation they had been on.
```
[NETWORK TASK COMPLETE: FIRST MULTI-TYPE COORDINATION]
[LP EARNED:]
[— MAO YINGJIE: 12,000 (CREDITED TO NETWORK NODE 2)]
[— LI XIULAN: 10,000 (CREDITED TO NETWORK NODE 3)]
[— CHEN HAORAN: 13,000 (CREDITED TO HOST)]
[TOTAL LP EARNED THIS TASK: 35,000]
[NOTE: COORDINATION EFFICIENCY — 71%. OPTIMAL EFFICIENCY FOR THIS TASK TYPE WAS 89%. THE PLANNING ERROR AND CORRECTION CYCLE COST 18 EFFICIENCY POINTS AND 4 DAYS.]
[NOTE: THE EFFICIENCY LOSS IS DOCUMENTED. THE TASK IS COMPLETE. BOTH MATTER.]
[CURRENT LP: 296,400]
```
He read the note twice.
Seventy-one percent. The system was tracking efficiency. The eighteen-point gap was the cost of the wrong model. The note said both completion and the efficiency loss mattered, which was accurate. The task was done. The planning error was also real.
He forwarded the completion notification to Mao Yingjie and Li Xiulan without comment.
Mao Yingjie responded in four minutes: *The node three call was wrong. I should have caught it in the initial analysis. I'll build the decision-criticality assessment into the pattern scan from the start next time.*
Li Xiulan responded in eleven minutes: *Noted. I'll flag model corrections within 24 hours of identification.*
He read both responses.
Three people who had found each other through a mechanism none of them asked for, running a coordinated task through a system none of them fully understood, making mistakes they were now documenting precisely so the same ones wouldn't appear in the next task.
The architect had designed this too, he thought. Not the mistakes — the capacity to correct them.
He put his phone down.
```
[NOTE: THE HEMING CAPITAL SITUATION HAS STABILIZED. THE LIN FAMILY FACTION'S SECONDARY FINANCIAL EXPOSURE IS UNCHANGED. THE CRISIS WINDOW DID NOT DEVELOP.]
[NOTE: LIN ZHENGYUE IS UNAWARE THAT A FINANCIAL PRESSURE POINT IN HER PLANNING HORIZON WAS REMOVED BY THE NETWORK'S INTERVENTION.]
[NOTE: THIS IS CORRECT OPERATION. PARAMETERS ARE YOURS.]
```
He looked at this for a long moment.
Correct operation. He had not told her. The financial pressure that would have complicated her Capital Alliance calculation in the same window as her Vanguard merger decision — it was gone now, resolved before it materialized, and she didn't know it had existed.
He thought about whether this was a problem.
It wasn't a deception. He hadn't told her he was monitoring her faction's financial exposure. He hadn't known he was until Mao Yingjie's pattern scan surfaced it. The network task had been about Heming Capital, not the Lin Family.
But the outcome benefited her, without her knowledge, through his action.
He filed it in the category that held the things that were true and complicated simultaneously. He had more than one of those. The category was getting used.
He opened the system shop and checked his LP balance.
296,400. Three hundred thousand within range. The next Unmarked Package unlocked at three hundred thousand.
Four days.
He closed the shop and went back to work.
